3.2.1. Nonlinear pattern classifiers
Taking up from the end of the previous works, the
features C and L=A were removed from the input data of the
classifier and so the results presented here refer to the input
made up of the four remaining features ða – e=A – R – PÞ:
In order to find the optimum number of neurons for the
intermediate layer of the nonlinear classifier, the empirical
criteria of gradually increasing the number of neurons in this
layer were used and consecutively the classification error
and performance were observed. This was carried out
considering the class of inclusion to be divided into linear
inclusion and nonlinear (a total of five classes), Fig. 3a, and
inclusion considered as only one class (total of four classes),
Fig. 3b. It was noted that in the first case, five classes, the
classifier reached a maximum performance (99.2%) and